Outlier Detector

Flag possible numeric outliers with an explicit IQR or z-score method while retaining every observation and its order.

On-device No sign-up Free to use

Find possible outliers controls

Showing an example. Edit to see your own.

?

Up to 5,000 numbers. Commas, spaces, semicolons or line breaks separate them.

?

The usual fence is 1.5. Larger values flag fewer rows.

?

A value is flagged when its absolute z-score is strictly larger than this threshold.

Processed in your browser. Your inputs stay on this device.

Result

Showing an example.

With values 1 through 8 and 100, use the 1.5 × IQR rule to flag 100 for review without deleting it.

1 of 9 values is a possible outlier (IQR method, 1.5 x IQR).

https://usefreetools.net/outlier-detector · 20 November 2026

Was this helpful?

How to use Outlier Detector

  1. Enter observations separated by commas, spaces, semicolons or line breaks. Keep the source order; flags are reported by original position.
  2. Choose IQR and a positive multiplier, or Z-score with a threshold and population/sample standard deviation. Only the selected method’s settings are used.
  3. Calculate and read the possible-outlier count and notes. IQR needs at least four values; z-score needs at least two.
  4. Download outlier-flags.csv for every observation. The table shows the first 50; IQR mode also offers outlier-box-plot.svg, with at most 200 flagged points drawn.

Example: Outlier Detector

With values 1 through 8 and 100, use the 1.5 × IQR rule to flag 100 for review without deleting it.

You add
Values (comma or line separated): 1,2,3,4,5,6,7,8,100 Method: Interquartile range (type 7 quartiles) IQR multiplier (k): 1.5 Z-score threshold: 3 Standard deviation: Population (divide by n)
You get
1 of 9 values is a possible outlier (IQR method, 1.5 x IQR).

Options

Detection rule
IQR compares values strictly outside Q1 − k×IQR and Q3 + k×IQR using type-7 quartiles. Z-score compares the absolute score with the threshold, so unusually low and high values can both be flagged. Boundary equality is not flagged.
Flagged observations
Review the original row or position with each flag. A flagged value stays part of the source and is not automatically corrected or deleted.

Supported inputs and limits

  • Up to 5,000 values, with at least four for IQR or two for z-score.
  • Values have magnitude at most 10¹⁵, up to 200 digits and exponents within ±1,000.
  • IQR multiplier is positive and at most 1,000,000; z threshold is 0 to 1,000,000.
  • IQR uses type-7 interpolated quartiles.
  • When IQR is zero, this implementation flags no values, even if a few observations differ from the repeated central values.
More details

Z-score uses the selected population or sample SD; a zero SD also produces no flags.

A flag does not prove an error, and no value is removed.

The first 50 observations appear in the table;

CSV retains all rows in original order.

The IQR chart draws at most 200 flagged points.

Z-score mode has numeric flags and CSV without a box plot.

Where your input is processed

This tool does not upload your inputs. See the limits for anything saved on your device.

An unusual value may still be valid

A flag can reflect a recording error, a real extreme event or a model that does not fit the data. Check units, collection conditions and the shape of the distribution before deciding what to do. Repeating the analysis with a different threshold does not by itself resolve the cause.

Questions about Outlier Detector

Does a flagged value get deleted?

No. All observations remain in their original order with an added flag. Review the context before changing a dataset.

Why can IQR and z-score flag different values?

They use different summaries. Extreme values affect the mean and SD used by z-scores, while IQR fences depend on quartiles. Match the method and threshold before comparing results.

Why are no values flagged when IQR is zero?

This implementation disables IQR flags when Q1 equals Q3. That can happen with repeated values even when another value differs, for example 1, 1, 1, 1, 100. No flags does not mean the data has no unusual observations.

Can I use z-scores on a constant list?

Ordinary z-scores divide by SD, which is zero for a constant list. The tool marks the no-spread case and flags nothing rather than treating its placeholder zero scores as evidence of normality.

Page updated · Project managed by Tony Hines · Report a problem with Outlier Detector